MODELING ACOUSTIC TRANSITIONS IN SPEECH BY MODIFIED HIDDEN MARKOV-MODELS WITH STATE DURATION AND STATE DURATION-DEPENDENT OBSERVATION PROBABILITIES

We propose a modified hidden Markov model (MHMM) that incorporates non parametric state duration and state duration-dependent observation pro babilities to reflect state transitions and to have accurate temporal structures in the HMM. In addition, to cope with the problem that resu lts from the use of insufficient amount of training data, we propose t o use the modified continuous density hidden Markov model (MCDHMM) wit h a different number of mixtures for the probabilities of state durati on-independent and state duration-dependent observation. We show that this proposed method yields improvement in recognition accuracy in com parison with the conventional CDHMM.